Hydromechanics parameter calculation method and preset expert model selection network training method

Abstract

The invention relates to the technical field of fluid mechanics, in particular to a fluid mechanics parameter calculation method and a preset expert model selection network training method. Obtaining a target research space; inputting the target research space into a preset expert model selection network, and outputting at least one target expert model corresponding to the target research space; and identifying the target research space by each target expert model, and calculating a target fluid mechanics parameter corresponding to the target research space according to an identification result. According to the method, the network is selected by using the preset expert model, and at least one most suitable target expert model can be matched for the specific target research space. Each target expert model optimizes the flow field type at which the target expert model is skilled, and the calculation results of a plurality of target experts are fused, so that the target fluid mechanics parameters can be identified and calculated more accurately. In addition, a high-complexity model can be prevented from being used for processing a simple flow field area, and unnecessary calculation amount is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of fluid mechanics, and specifically relates to a method for calculating fluid mechanics parameters and a method for training a preset expert model selection network. Background Art

[0002] Computational Fluid Dynamics (CFD), as an important means for studying fluid flow problems, has been widely applied in many fields such as aerospace, automotive engineering, energy power, and building design. In CFD simulations, in order to accurately simulate various flow phenomena and obtain accurate results, it is crucial to select appropriate numerical simulation methods and improve grid accuracy. However, the current technological development in the CFD field faces many challenges.

[0003] In the field of Computational Fluid Dynamics (CFD), in order to accurately simulate different flow phenomena, it is usually necessary to select appropriate numerical simulation methods and improve the accuracy of the grid as much as possible to obtain accurate results. In traditional methods, regional partition calculation methods and adaptive grid refinement methods are usually used. Although they play a certain role in calculating fluid mechanics parameters, traditional methods not only lead to a large consumption of computing resources, and the computing efficiency is difficult to meet the actual engineering needs, but also have deficiencies in computing accuracy and stability.

[0004] Therefore, how to complete efficient, accurate, and stable calculation of fluid mechanics parameters has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the present invention provides a method for calculating fluid mechanics parameters and a method for training a preset expert model selection network to solve the problem of how to complete efficient, accurate, and stable calculation of fluid mechanics parameters.

[0006] In the first aspect, the present invention provides a method for calculating fluid mechanics parameters, and the method includes:

[0007] Obtain a target research space;

[0008] Input the target research space into a preset expert model selection network, and output at least one target expert model corresponding to the target research space;

[0009] Each target expert model identifies the target research space, and calculates the target fluid mechanics parameters corresponding to the target research space according to the identification result.

[0010] The hydrodynamic parameter calculation method provided by the embodiments of the present application obtains a target research space. The target research space is input into a preset expert model selection network, and at least one target expert model corresponding to the target research space is output, ensuring the accuracy of at least one target expert model corresponding to the output target research space and its matching with the target research space, and avoiding invalid calculations using unsuitable models. For example, when simulating aircraft flight, different models suitable for the complex flow field near the wing and the simple flow field far from the fuselage can be quickly selected. Compared with single-model calculation, the calculation time is significantly reduced and the calculation efficiency is improved. Each target expert model identifies the target research space and calculates the target hydrodynamic parameters corresponding to the target research space according to the identification results. Different target research spaces have unique hydrodynamic characteristics, and it is difficult for a single model to accurately simulate them. The above method uses a preset expert model selection network to match the most suitable at least one target expert model for a specific target research space. Each target expert model is optimized for the flow field type it is good at, and the calculation results of multiple target experts are fused to more accurately identify and calculate the target hydrodynamic parameters. In addition, the above method selects a suitable target expert model through a preset expert model selection network, which can avoid using high-complexity models to process simple flow field regions and reduce unnecessary calculation amounts. At the same time, the model adapted to the target research space can adopt a more reasonable grid accuracy to further optimize the use of computing resources. When simulating the wind environment of large buildings, a reasonable selection of models and grids can significantly reduce computing resource consumption while ensuring calculation accuracy. In practical engineering applications, the working conditions of the target research space are complex and changeable. The preset expert model selection network of this method can flexibly select target expert models according to different target research space characteristics, or even combine multiple models to work together, so as to ensure the accuracy of calculation results while reducing computing resources. When studying the flow field of an aircraft in different flight postures, the models can be quickly switched or combined to adapt to the changes in working conditions, improve the adaptability and flexibility of the models, and meet diverse engineering needs. The above method completes efficient, accurate and stable hydrodynamic parameter calculation.

[0011] In an alternative embodiment, inputting the target research space into a preset expert model selection network and outputting at least one target expert model corresponding to the target research space includes:

[0012] Dividing the target research space by region to obtain a plurality of target regions; [[ID=G]]

[0013] Inputting each target region corresponding to the target research space into the preset expert model selection network;

[0014] For each target region, outputting at least one target expert model corresponding to the target region.

[0015] The hydrodynamic parameter calculation method provided by the embodiments of the present application divides the target research space into regions to obtain multiple target regions, enabling a more detailed analysis of the characteristics of each target region. Each target region corresponding to the target research space is input into a preset expert model selection network; for each target region, at least one target expert model corresponding to the target region is output. Thus, the target expert model can accurately simulate the flow characteristics of a specific target region, greatly improving the calculation accuracy, and can match the complexity of the target region with the target expert model. A model with low computational cost can be used for simple regions, and a high-precision model can be used for complex regions, rationally allocating computational resources and reducing unnecessary computational volume, thereby improving the calculation efficiency. In addition, the situation of the target research space in practical applications is complex and changeable, and a single model is difficult to adapt to the flow characteristics of all regions. The method of dividing regions and selecting models enhances the adaptability of the model. When the boundary conditions or internal structure of the target research space change, for the target experts of the machine learning type, only the model selection of the affected region needs to be adjusted, without having to recalculate the entire space. When simulating the impact of river diversion on the flow field of the surrounding waters, only the model calculation for the diversion region and the affected surrounding regions needs to be reselected, which can quickly respond to changes and adapt to different working conditions.

[0016] In an alternative embodiment, for each target region, outputting at least one target expert model corresponding to the target region includes:

[0017] For each target region, the preset expert model selection network extracts features of the target region to obtain target parameter features corresponding to the target region;

[0018] Based on the target parameter features, at least one target expert model corresponding to the target region is output.

[0019] The hydrodynamic parameter calculation method provided by the embodiments of the present application, for each target region, the preset expert model selection network extracts features of the target region to obtain target parameter features corresponding to the target region; different target regions have unique hydrodynamic characteristics. The preset expert model selection network extracts features of the target region, which can accurately capture these characteristics and convert them into target parameter features, ensuring the accuracy of the extracted target parameter features. Then, based on the target parameter features, at least one target expert model corresponding to the target region is output, ensuring that each output target expert model matches the target region.

[0020] In an alternative embodiment, each target expert model identifies the target research space, and calculates the target hydrodynamic parameters corresponding to the target research space according to the identification result, including:

[0021] For each target area, input the target research space into each target expert model corresponding to the target area, and the target expert model outputs the sub-target hydrodynamic parameters corresponding to the target area;

[0022] Fuse the sub-target hydrodynamic parameters to output the regional hydrodynamic parameters corresponding to the target area;

[0023] Fuse the regional hydrodynamic parameters corresponding to each target area to generate the target hydrodynamic parameters corresponding to the target research space.

[0024] In the hydrodynamic parameter calculation method provided by the embodiments of the present application, for each target area, the target research space is input into each target expert model corresponding to the target area, and the target expert model outputs the sub-target hydrodynamic parameters corresponding to the target area. The hydrodynamic characteristics of different target areas vary greatly. Each target expert model simulates the flow characteristics of a specific target area and can more accurately reflect the physical phenomena in the specific target area, thereby ensuring the accuracy of the sub-target hydrodynamic parameters corresponding to the output target area. Then, fuse the sub-target hydrodynamic parameters to output the regional hydrodynamic parameters corresponding to the target area. This avoids the accumulation of errors that may occur when a single model simulates the entire target research space in different regions. Within each target area, the calculations of each target expert model are relatively independent, reducing the propagation of errors. Moreover, during the fusion process, comprehensive processing is performed on the sub-target hydrodynamic parameters, further improving the accuracy of the regional hydrodynamic parameters corresponding to the target area. Finally, fuse the regional hydrodynamic parameters corresponding to each target area to generate the target hydrodynamic parameters corresponding to the target research space, ensuring the accuracy of the generated target hydrodynamic parameters.

[0025] In a second aspect, the present invention provides a method for training a preset expert model selection network, which is applied to the preset expert model selection network in the first aspect or any corresponding embodiment thereof. The method includes:

[0026] Obtain a training data set, which includes each training research space and the training hydrodynamic parameters corresponding to each training research space under various working conditions;

[0027] Input each training research space into each training expert model included in the initial expert model selection network, and each training expert model outputs the virtual hydrodynamic parameters corresponding to each training research space under various working conditions;

[0028] Based on the correspondence between the virtual hydrodynamic parameters and the training hydrodynamic parameters, adjust the parameters of the initial expert model selection network to obtain a preset expert model selection network.

[0029] The preset expert model selection network training method provided by the embodiment of the present application obtains a training data set. The training data set covers the real fluid mechanics parameters of various training research spaces under different working conditions. Each training research space is input into each training expert model included in the initial expert model selection network, and each training expert model outputs the virtual fluid mechanics parameters corresponding to each training research space under various working conditions. Based on the corresponding relationship between each virtual fluid mechanics parameter and each training fluid mechanics parameter, the parameters of the initial expert model selection network are adjusted to obtain a preset expert model selection network. By continuously adjusting the network parameters, the preset expert model selection network can accurately select the target model most suitable for simulating the fluid mechanics phenomenon of the space according to the characteristics and working conditions of the training research space, thereby improving the accuracy of the selection of the preset expert model selection network.

[0030] In an alternative embodiment, the training research space includes at least one training area; each training area is obtained by dividing the training research space by area; the training fluid mechanics parameters include the sub-training fluid mechanics parameters corresponding to each training area; inputting each training research space into each training expert model included in the initial expert model selection network, and each training expert model outputs the virtual fluid mechanics parameters corresponding to each training research space under various working conditions, including:

[0031] Input each training research space into each training expert model included in the initial expert model selection network;

[0032] For each training area in the training research space, each training expert model extracts features from each training area and outputs the training parameter features corresponding to each training area;

[0033] Based on each training parameter feature, output the sub-virtual fluid mechanics parameters corresponding to each training area under various working conditions.

[0034] The preset expert model selection network training method provided by the embodiments of the present application inputs each training research space into each training expert model included in the initial expert model selection network. For each training area in the training research space, each training expert model extracts features from each training area and outputs the training parameter features corresponding to each training area. Thus, each training expert model can capture the accurate features corresponding to the training area for each training area, ensuring the accuracy of the output training parameter features. In addition, the regional processing makes the training of each training area relatively independent, avoiding the interference of the complex interaction between different training areas on the training results. Then, based on the training parameter features, the sub-virtual fluid mechanics parameters corresponding to each training area under various working conditions are output, ensuring the accuracy of the sub-virtual fluid mechanics parameters corresponding to each training area under various working conditions. The above method processes each training area separately, and can reasonably allocate computing resources according to the complexity and computing requirements of each area. In addition, the regional training can be carried out in parallel, and the multi-core computing resources are used to train multiple training areas simultaneously, improving the training efficiency.

[0035] In an alternative embodiment, based on the correspondence between each virtual fluid mechanics parameter and each training fluid mechanics parameter, the parameters of the initial expert model selection network are adjusted to obtain a preset expert model selection network, including:

[0036] For each training area, based on the attribute information corresponding to each training area, at least one evaluation mechanism corresponding to each training area is determined;

[0037] Based on the objective function corresponding to each evaluation mechanism, the evaluation scores corresponding to each training expert model under various working conditions are calculated;

[0038] According to the evaluation scores corresponding to each training expert model under various working conditions, at least one candidate expert model is determined therefrom;

[0039] Based on each candidate expert model, the parameters of the initial expert model selection network are adjusted to obtain a preset expert model selection network.

[0040] The preset expert model selection network training method provided by the embodiments of the present application determines at least one evaluation mechanism corresponding to each training area based on the attribute information corresponding to each training area for each training area, so as to more accurately measure the performance of the training expert model in a specific training area. Based on the objective functions corresponding to each evaluation mechanism, calculate the evaluation scores corresponding to each training expert model under various working conditions, so as to comprehensively evaluate the performance of each training expert model in different working conditions for each training area. According to the evaluation scores corresponding to each training expert model under various working conditions, determine at least one candidate expert model therefrom, so that the initial expert model selection network is adjusted based on the candidate expert model, so that the initial expert model selection network is more inclined to select a training expert model with excellent performance. Adjust the parameters of the initial expert model selection network based on each candidate expert model to obtain a preset expert model selection network, ensuring the accuracy of the obtained preset expert model selection network.

[0041] In an alternative embodiment, adjusting the parameters of the initial expert model selection network based on each candidate expert model to obtain a preset expert model selection network includes:

[0042] For each training area, perform fusion processing on the sub-virtual hydrodynamics parameters output by each candidate expert model corresponding to the training area to obtain a first fusion hydrodynamics parameter;

[0043] Calculate the loss function between the first fusion hydrodynamics parameter corresponding to each training area and the corresponding sub-training hydrodynamics parameter;

[0044] Fuse the loss functions corresponding to each training area to generate an objective loss function;

[0045] According to the objective loss function, adjust the parameters of the initial expert model selection network until the objective loss function stabilizes within a preset area to obtain a preset expert model selection network.

[0046] The preset expert model selection network training method provided by the embodiments of this application performs fusion processing on the sub-virtual hydrodynamics parameters output by each candidate expert model corresponding to a training area for each training area to obtain the first fused hydrodynamics parameter, which can integrate the advantages of multiple candidate expert models, make up for the limitations of a single candidate expert model, and ensure the accuracy of the obtained first fused hydrodynamics parameter. Calculate the loss function between the first fused hydrodynamics parameter corresponding to each training area and the corresponding sub-training hydrodynamics parameter, and fuse the loss functions corresponding to each training area to generate the target loss function, which provides a clear optimization target for the parameter adjustment of the initial expert model selection network. Adjusting the parameters of the initial expert model selection network according to the target loss function enables the initial expert model selection network to gradually learn how to select and combine models to minimize the difference between the prediction result and the true value. As the parameters are continuously adjusted, the prediction accuracy of the preset expert model selection network will gradually improve, and finally, it can more accurately simulate the hydrodynamics characteristics of different training areas. In addition, adjusting the network parameters based on the loss functions of each training area enables the preset expert model selection network to better adapt to the characteristics of different training areas and the changes in various working conditions. Each training area has its unique hydrodynamics characteristics and boundary conditions. By calculating and fusing the loss functions separately, the preset expert model selection network can perform personalized optimization for different areas, improving the adaptability of the preset expert model selection network in different areas. Considering the changes in hydrodynamics phenomena under different working conditions, this method can also enable the model to maintain good performance under various working conditions, enhancing the reliability of the model.

[0047] In a third aspect, the present invention provides a hydrodynamics parameter calculation device, which includes:

[0048] A first acquisition module for acquiring a target research space;

[0049] An output module for inputting the target research space into a preset expert model selection network and outputting at least one target expert model corresponding to the target research space;

[0050] A calculation module for each target expert model to identify the target research space and calculate the target hydrodynamics parameter corresponding to the target research space according to the identification result.

[0051] The hydrodynamic parameter calculation device provided by the embodiment of the present application obtains a target research space. The target research space is input into a preset expert model selection network, and at least one target expert model corresponding to the target research space is output, ensuring the accuracy of at least one target expert model corresponding to the output target research space and its matching with the target research space, and avoiding invalid calculations using inappropriate models. For example, when simulating aircraft flight, different models suitable for the complex flow field near the wing and the simple flow field far from the fuselage can be quickly selected. Compared with single-model calculation, the calculation time is significantly reduced and the calculation efficiency is improved. Each target expert model identifies the target research space and calculates the target hydrodynamic parameters corresponding to the target research space according to the identification result. Different target research spaces have unique hydrodynamic characteristics, and it is difficult for a single model to accurately simulate them. The above device uses a preset expert model selection network to match the most suitable at least one target expert model for a specific target research space. Each target expert model is optimized for the flow field type it is good at, and the calculation results of multiple target experts are fused, which can more accurately identify and calculate the target hydrodynamic parameters. In addition, the above device selects a suitable target expert model through the preset expert model selection network, which can avoid using high-complexity models to process simple flow field regions and reduce unnecessary calculation amounts. At the same time, a grid with slightly lower accuracy can be used for the model adapted to the target research space to further optimize the use of computing resources. When simulating the wind environment of large buildings, reasonable selection of models and grids can significantly reduce computing resource consumption on the premise of ensuring calculation accuracy. In practical engineering applications, the working conditions of the target research space are complex and changeable. The preset expert model selection network of this device can flexibly select target expert models according to different target research space characteristics, and even combine multiple models to work together, so as to ensure the accuracy of calculation results while reducing computing resources. When studying the flow field of an aircraft in different flight postures, the model can be quickly switched or combined to adapt to the change of working conditions, improve the adaptability and flexibility of the model, and meet diverse engineering needs.

[0052] In a fourth aspect, the present invention provides a training device for a preset expert model selection network, which is applied to the preset expert model selection network in the first aspect or any corresponding embodiment thereof. The device includes:

[0053] A second acquisition module, configured to acquire a training data set, where the training data set includes each training research space and the training hydrodynamic parameters corresponding to each training research space under various working conditions;

[0054] An input module, configured to input each training research space into each training expert model included in the initial expert model selection network, and each training expert model outputs the virtual hydrodynamic parameters corresponding to each training research space under various working conditions;

[0055] An adjustment module is configured to adjust the parameters of the initial expert model selection network based on the correspondence between each virtual fluid dynamics parameter and each training fluid dynamics parameter, so as to obtain a preset expert model selection network.

[0056] The preset expert model selection network training device provided by the embodiments of the present application obtains a training data set. The training data set covers the real fluid dynamics parameters of various training research spaces under different working conditions. Each training research space is input into each training expert model included in the initial expert model selection network, and each training expert model outputs the virtual fluid dynamics parameters corresponding to each training research space under various working conditions. Based on the correspondence between each virtual fluid dynamics parameter and each training fluid dynamics parameter, the parameters of the initial expert model selection network are adjusted to obtain a preset expert model selection network. By continuously adjusting the network parameters, the preset expert model selection network can accurately select the target model most suitable for simulating the fluid dynamics phenomenon of the space according to the characteristics and working conditions of the training research space, thereby improving the accuracy of the selection of the preset expert model selection network. Description of the Drawings

[0057] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0058] Figure 1 It is a flowchart of a fluid dynamics parameter calculation method according to an embodiment of the present invention;

[0059] Figure 2 It is a flowchart of another fluid dynamics parameter calculation method according to an embodiment of the present invention;

[0060] Figure 3 It is a flowchart of a preset expert model selection network training method according to an embodiment of the present invention;

[0061] Figure 4 It is a structural block diagram of a fluid dynamics parameter calculation device according to an embodiment of the present invention;

[0062] Figure 5 It is a structural block diagram of a preset expert model selection network training device according to an embodiment of the present invention;

[0063] Figure 6 It is a hardware structural schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Embodiments

[0064] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0065] Computational Fluid Dynamics (CFD), as an important means for studying fluid flow problems, has been widely applied in many fields such as aerospace, automotive engineering, energy and power, and building design. In CFD simulations, to accurately simulate various flow phenomena and obtain accurate results, it is crucial to select appropriate numerical simulation methods and improve the grid accuracy. However, the current technological development in the CFD field faces many challenges.

[0066] In the field of Computational Fluid Dynamics (CFD), to accurately simulate different flow phenomena, it is usually necessary to select suitable numerical simulation methods and improve the grid accuracy as much as possible to obtain accurate results. In traditional methods, the regional partition calculation method and the adaptive grid refinement method are usually adopted. Although they play a certain role in the calculation of fluid mechanics parameters, traditional methods not only consume a large amount of computing resources and the computing efficiency is difficult to meet the actual engineering needs, but also have deficiencies in terms of calculation accuracy and stability.

[0067] Therefore, how to complete the calculation of fluid mechanics parameters efficiently, accurately, and stably has become an urgent problem to be solved.

[0068] It should be noted that the method for calculating fluid mechanics parameters provided in the embodiments of the present application may be executed by a device for calculating fluid mechanics parameters. The device for calculating fluid mechanics parameters can be implemented as part or all of an electronic device through software, hardware, or a combination of software and hardware. Among them, the electronic device can be a server or a terminal. Among them, the server in the embodiments of the present application can be a single server or a server cluster composed of multiple servers. The terminal in the embodiments of the present application can be other intelligent hardware devices such as a smart phone, a personal computer, a tablet computer, a wearable device, and a smart robot. In the following method embodiments, the execution subject is taken as an electronic device for illustration.

[0069] According to the embodiments of the present invention, an embodiment of a method for calculating fluid mechanics parameters is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0070] In this embodiment, a method for calculating hydrodynamic parameters is provided, which can be used in the above-mentioned electronic device. Figure 1 It is a flowchart of the method for calculating hydrodynamic parameters according to an embodiment of the present invention, as Figure 1 shown. The process includes the following steps:

[0071] Step S101, obtain the target research space.

[0072] Specifically, the electronic device can receive the target research space input by the user or the target research space sent by other devices. Among them, the target research space can be a wind tunnel experiment space or a water tunnel experiment space. Among them, the wind tunnel experiment space is an important place for studying aerodynamics in the fields of aerospace, automotive, etc. In the wind tunnel, stable airflows are generated by devices such as fans to simulate the air flow conditions of objects at different speeds and angles. Researchers can place aircraft models, car models, etc. in this space to measure hydrodynamic parameters such as the pressure distribution, air flow velocity, lift, and drag on the model surface, so as to optimize the design and improve the performance of aircraft or the fuel economy of cars, etc. The water tunnel experiment space is mainly used to study the hydrodynamic characteristics of objects in water, such as the navigation performance of ships, the movement of underwater robots, etc. The water tunnel can simulate different water flow velocities, flow directions, and turbulent states of the water flow. By conducting experiments in the water tunnel, data such as the pressure, friction force, and wake characteristics on the object surface can be obtained, providing an important basis for ship design and the research and development of underwater equipment.

[0073] In addition, the target research space can also be the interior space or exterior space of a building. Among them, the interior space of a building: such as the indoor spaces of large shopping malls, stadiums, office buildings, etc. When designing these buildings, factors such as indoor air flow, temperature distribution, and ventilation effect need to be considered to provide a comfortable indoor environment. By conducting hydrodynamic analysis on the interior space of the building, simulating the air flow organization under working conditions such as personnel activities and the operation of the air conditioning system, and optimizing the positions of ventilation openings and the forms of air conditioning outlets, etc., the indoor air quality and thermal comfort can be improved. The exterior space of a building is the wind environment around the building. Studying the air flow distribution around the building under different wind directions and wind speeds is of great significance for evaluating the wind load of the building, pedestrian comfort, and pollutant diffusion, etc. For example, in urban planning, reasonably designing the layout and shape of buildings can reduce the impact of strong winds on pedestrians and avoid damage to buildings caused by excessive wind pressure differences in local areas.

[0074] In addition, the target research space can also be the atmospheric boundary layer space and the river-lake space. Among them, the atmospheric boundary layer space is the transition layer between the Earth's surface and the free atmosphere, with a height generally around 1000 to 2000 meters. Studying processes such as the wind field, temperature field, humidity field, and the diffusion and transmission of pollutants within the atmospheric boundary layer is of great significance for weather forecasting, air quality assessment, climate change research, etc. By establishing an atmospheric boundary layer model and combining meteorological observation data, the diffusion law of pollutants under different weather conditions can be simulated, providing a scientific basis for environmental management and pollution control. The river-lake space is the space for studying the water flow movement, water quality changes, and ecosystem in rivers, lakes and other water bodies. For example, analyzing the velocity distribution and water level changes of rivers is of great significance for flood control and disaster reduction, and water resources allocation; studying the eutrophication process and pollutant diffusion law in lakes is crucial for protecting the lake ecological environment and preventing water pollution. By establishing hydrodynamic models and water quality models and numerically simulating the river-lake space, the changing trend of water bodies can be predicted, providing decision-making support for water environment governance and ecological restoration.

[0075] Finally, the target research space can also be other spaces, and the embodiments of the present application do not make specific limitations on the target research space.

[0076] Step S102: Input the target research space into a preset expert model selection network, and output at least one target expert model corresponding to the target research space.

[0077] Specifically, the electronic device inputs the target research space into the input layer of the preset expert model selection network. The hidden layer in the preset expert model selection network extracts features from the target research space. For example, the convolutional layer in the CNN performs convolution operations by sliding the convolution kernel over the data to extract local features in the data. The RNN and its variants can process data with sequence information and capture the time series features in the data. Through the hierarchical processing of multiple hidden layers, more advanced and abstract features are gradually extracted.

[0078] Among them, the preset expert model selection network includes strategies and algorithms for model selection. Optionally, the target research space can be based on the attention mechanism, and by calculating the importance weights of each expert model for the current target research space, expert models can be dynamically selected and combined. Optionally, the preset expert model selection network can also adopt a gating mechanism to determine which expert models to activate according to the features of the input target research space.

[0079] Finally, after the processing inside the preset expert model selection network, the result is obtained at the output layer. The number of neurons in the output layer corresponds to the number of expert models, and the value of each neuron represents the probability or score of the corresponding expert model being selected.

[0080] According to the results of the output layer, select one or more expert models with the highest score or the highest probability as the target expert models. For example, if there are three neurons in the output layer, corresponding to three expert models A, B, and C respectively, and their output values are 0.6, 0.3, and 0.1 respectively, then expert model A can be selected as the target expert model. If multiple target expert models need to be selected, a threshold can be set to select the target expert models with output values greater than the threshold.

[0081] This step will be introduced in detail below.

[0082] Step S103: Each target expert model identifies the target research space and calculates the target hydrodynamic parameters corresponding to the target research space according to the identification results.

[0083] Among them, the target expert model can be a model of computational theory methods, such as the finite volume method (FVM), lattice Boltzmann method (LBM), smoothed particle hydrodynamics (SPH), etc.; it can also be a physical model, such as RANS - type models (standard k - ε model, SST k - ω model, etc.), LES - type models (Smagorisky model, dynamic Smagorisky model), DES - type hybrid models, etc.; it can also be a traditional adaptive scheme model, such as the regional partition calculation method, adaptive grid refinement method, multigrid method, etc.; in addition, it can also be an artificial neural network (Artificial Neural Network, ANN) model, support vector machine (Support Vector Machine, SVM) model, reduced order model (Reduced Order Model, ROM), physics - informed neural network (Physics - Informed Neural Network, PINN) model. The embodiments of the present application do not make specific limitations on the target expert model.

[0084] Among them, the Reduced Order Model (ROM) combines the prior knowledge of physical models and data-driven methods. By reducing the dimension of high-dimensional physical models, it extracts key physical information and features and establishes a low-dimensional model to approximately describe the behavior of the original system. It is commonly used in fluid mechanics for rapid prediction of the dynamic response of flow fields, optimization design, etc. For example, in the design of aeroengines, it is used to quickly evaluate the influence of different design parameters on the performance of the flow field. The Physics-Informed Neural Network (PINN) model embeds physical laws (such as the Navier-Stokes equations, etc.) into the neural network in the form of constraints, enabling the neural network to follow physical laws while learning data, thereby improving the accuracy and generalization ability of the model. In fluid mechanics, it can be used to solve partial differential equations, predict the evolution of complex flow fields, etc., such as simulating fluid flow problems with complex boundary conditions and physical processes.

[0085] Specifically, the target expert model first identifies the geometry of the target research space. Taking the simulation of the flow field around a car as an example, the model obtains geometric information such as the outer contour of the car, the body size, the shapes and positions of various components. Through this information, the model can determine the flow paths and boundary conditions of the fluid at different parts, providing a basis for subsequent calculations.

[0086] Then, the target expert model identifies the boundary conditions of the target research space. Among them, the boundary conditions include the flow velocity, temperature, and pressure at the inlet, the pressure conditions at the outlet, and the roughness and temperature of the wall surface. When simulating the ventilation system inside a building, the target expert model needs to identify boundary conditions such as the wind speed and temperature at the ventilation openings and the heat conduction characteristics of the building wall surface. These conditions directly affect the distribution of the flow field and the calculation of fluid mechanics parameters.

[0087] In addition, the target expert model also identifies the physical properties of the fluid within the target research space, such as density, viscosity, and thermal conductivity. When simulating the flow of different fluids (such as water, air, oil, etc.) in a specific space, the accurate identification of these physical properties is crucial for the accuracy of the calculation results.

[0088] Finally, based on the recognition results, the target expert model begins to calculate the target hydrodynamic parameters. The most basic parameters include flow velocity, pressure, and density. When simulating the fluid flow in a pipeline, the model will solve the flow velocity distribution and pressure distribution of the fluid through numerical calculation methods according to the geometric shape of the pipeline, boundary conditions, and physical properties of the fluid. For example, the finite element method or the finite volume method is used to discretize the target research space, and then the hydrodynamic equations are applied to each discrete element for calculation. In addition to the basic parameters, some complex hydrodynamic parameters will also be calculated, such as vorticity, shear stress, turbulence intensity, etc. When simulating the flow in the atmospheric boundary layer, the calculation of vorticity can help understand the formation and development of vortices in the atmosphere; the calculation of shear stress is of great significance for analyzing the frictional force and energy loss of the fluid near the wall; the calculation of turbulence intensity can evaluate the turbulence degree of the flow field and provide a basis for further analysis and prediction. In some complex target research spaces, there are also couplings of multiple physical fields, such as fluid-thermal coupling, fluid-solid coupling, etc. When simulating the working process of an engine combustion chamber, not only the fluid flow but also factors such as chemical reactions, heat transfer, and thermal deformation of the solid wall during the combustion process need to be considered. The target expert model needs to comprehensively consider the interactions of these multiple physical fields and calculate the corresponding coupling parameters, such as temperature distribution, heat flux, structural stress, etc.

[0089] The hydrodynamic parameter calculation method provided by the embodiments of the present application obtains a target research space. The target research space is input into a preset expert model selection network, and at least one target expert model corresponding to the target research space is output, ensuring the accuracy of at least one target expert model corresponding to the output target research space and its matching with the target research space, and avoiding invalid calculations using inappropriate models. For example, when simulating aircraft flight, a low-order model applicable to the complex flow field near the wing and the simple flow field far from the fuselage can be quickly selected, significantly reducing the calculation time and improving the calculation efficiency. Each target expert model identifies the target research space and calculates the target hydrodynamic parameters corresponding to the target research space according to the identification result. Different target research spaces have unique hydrodynamic characteristics, and it is difficult for a single model to accurately simulate them. The above method uses a preset expert model selection network to match the most suitable at least one target expert model for a specific target research space. Each target expert model is optimized for the flow field type it is good at, and the calculation results of multiple target experts are fused to more accurately identify and calculate the target hydrodynamic parameters. In addition, the above method selects a suitable target expert model through a preset expert model selection network, which can avoid using a high-complexity model to process simple flow field areas and reduce unnecessary calculation amounts. At the same time, a more reasonable grid accuracy can be adopted for the model adapted to the target research space to further optimize the use of computing resources. When simulating the wind environment of a large building, a reasonable selection of the model and grid can significantly reduce the consumption of computing resources while ensuring the calculation accuracy. In practical engineering applications, the working conditions of the target research space are complex and variable. The preset expert model selection network of this method can flexibly select target expert models according to different target research space characteristics, or even combine multiple models to work together, so as to ensure the accuracy of the calculation results while reducing the computing resources. When studying the flow field of an aircraft in different flight postures, the model can be quickly switched or combined to adapt to the change of working conditions, improve the adaptability and flexibility of the model, and meet diverse engineering needs. The above method completes the calculation of hydrodynamic parameters efficiently, accurately and stably.

[0090] In this embodiment, a hydrodynamic parameter calculation method is provided, which can be used in the above-mentioned electronic device. Figure 2 It is a flowchart of the hydrodynamic parameter calculation method according to the embodiments of the present invention, as Figure 2 shown. The process includes the following steps:

[0091] Step S201, obtain a target research space.

[0092] For this step, please refer to the introduction of step S101 above and will not be elaborated here.

[0093] Step S202: Input the target research space into a preset expert model selection network, and output at least one target expert model corresponding to the target research space.

[0094] Specifically, the above Step S202 may include the following steps:

[0095] Step S2021: Divide the target research space into regions to obtain multiple target regions.

[0096] Specifically, the electronic device may divide the target research space into regions according to the physical characteristics corresponding to the target research space to obtain multiple target regions. Among them, the physical characteristics may include fluid state characteristics, temperature characteristics, pressure characteristics, etc.

[0097] Exemplarily, in fluid mechanics research, the electronic device may divide the regions according to the flow state of the target research space. For example, the target research space may be divided into a laminar flow region and a turbulent flow region. In pipe flow, the region near the pipe wall may be in a laminar state due to low flow velocity and large viscous action, while the central region of the pipe has a high flow velocity and may form turbulence. Through this division, different models and methods can be used to study the fluid characteristics under different flow states. For a target research space involving heat transfer, the regions can be divided according to the temperature distribution. For example, when studying the fluid temperature change in a heat exchanger, the region with a relatively gentle temperature change can be divided into one target region, and the region with an obvious temperature gradient can be divided into another target region, so as to more specifically analyze the heat transfer process and fluid flow characteristics in different temperature regions. Divide according to the pressure distribution in the target research space. When studying the flow field around an aircraft wing, the pressure distribution on the wing surface is uneven. The regions with higher pressure and lower pressure can be divided into different target regions respectively to analyze the influence of the pressure difference on the lift and drag of the wing.

[0098] Optionally, the electronic device may divide the target research space into regions according to the function or application requirements corresponding to the target research space to obtain multiple target regions.

[0099] Exemplarily, the target research space is the interior space of a building. The electronic device can divide the interior space of the building into different functional areas, such as living areas, public activity areas, equipment room areas, etc. For each functional area, the requirements for air flow organization, temperature control, etc. may be different. Through this division, the design and analysis of systems such as ventilation and air conditioning can be carried out in a targeted manner. For example, in a chemical reactor, the target research space is the space inside the reactor. The electronic device can divide the space inside the reactor into a feed area, a reaction area, a discharge area, etc. according to the reaction process and material flow conditions. This can better study the chemical reaction rate, material mixing situation, etc. in different areas, and optimize the production process and equipment design. For a large ecological system research space, it can be divided into different target areas according to ecological functions, such as habitat protection areas, buffer zones, experimental research areas, etc. This division helps to formulate different protection and research strategies to protect the ecological environment and biodiversity.

[0100] Step S2022: Input each target area corresponding to the target research space into a preset expert model selection network.

[0101] Specifically, the electronic device can input each target area corresponding to the target research space into a preset expert model selection network.

[0102] Exemplarily, assume that the target research space is the interior space of a building, and assume that the electronic device divides the interior space of the building into 4 target areas. Then the electronic device inputs the 4 target areas corresponding to the interior space of the building into the preset expert model selection network according to the positional relationship before division.

[0103] Step S2023: Output at least one target expert model corresponding to each target area.

[0104] Specifically, the above step S2023 may include the following steps:

[0105] Step a1: For each target area, the preset expert model selection network extracts features of the target area to obtain target parameter features corresponding to the target area.

[0106] Specifically, for each target area, the preset expert model selection network extracts features of the target area to obtain target parameter features corresponding to the target area.

[0107] Exemplarily, for a target region with a geometric shape, the preset expert model selection network can extract the area, perimeter, volume, curvature, etc. corresponding to the target region. When simulating the water flow in a river bend (target region), the radius of curvature of the bend is an important geometric feature, which will affect the velocity distribution and pressure change of the water flow. For a target region involving fluids, the property characteristics such as the density, viscosity, specific heat capacity, and thermal conductivity of the fluid can be extracted. When studying the heat transfer and flow problems in a certain channel (target region) of a heat exchanger, the thermal conductivity and specific heat capacity of the fluid will affect the heat transfer rate, while the viscosity is related to the flow resistance of the fluid. If the target region contains a solid part, the property characteristics such as the elastic modulus, Poisson's ratio, and thermal expansion coefficient of the solid can be extracted. In fluid-structure interaction problems, such as the pipeline vibration analysis caused by the fluid flow in a pipeline, the characteristics such as the elastic modulus and Poisson's ratio of the solid pipeline will affect the deformation and vibration characteristics of the pipeline. In addition, the inlet boundary characteristics and wall boundary characteristics corresponding to the target region can also be extracted. For example, the velocity magnitude, direction, temperature, pressure, etc. at the inlet of the target region are extracted. When simulating the air flow in an air duct, the wind speed and temperature at the inlet of the air duct are the key factors affecting the air flow and heat exchange in the air duct. For the wall of the target region, the roughness, temperature, heat flux density, friction force, etc. of the wall are extracted. When studying the flow of a fluid in a pipeline, the roughness of the pipeline wall will affect the flow resistance and turbulence characteristics of the fluid.

[0108] The embodiments of the present application do not specifically limit the target parameter characteristics corresponding to the target region.

[0109] Step a2, based on the target parameter characteristics, output at least one target expert model corresponding to the target region.

[0110] Specifically, the preset expert model selection network identifies the target parameter characteristics and outputs at least one target expert model corresponding to the target region.

[0111] Step S203, each target expert model identifies the target research space and calculates the target fluid mechanics parameters corresponding to the target research space according to the identification result.

[0112] Specifically, the above step S203 may include the following steps:

[0113] Step S2031, for each target region, input the target research space into each target expert model corresponding to the target region, and the target expert model outputs the sub-target fluid mechanics parameters corresponding to the target region.

[0114] Specifically, for each target region, the electronic device inputs the target research space into each target expert model corresponding to the target region, and the target expert model outputs the sub-target fluid mechanics parameters corresponding to the target region.

[0115] Exemplarily, assume that the target research space is the interior space of a building, and assume that the electronic device divides the interior space of the building into 4 target areas. For the first target area, the electronic device inputs the target research space into each target expert model corresponding to the first target area. The target expert model outputs the sub-target fluid mechanics parameters corresponding to the first target area. Similarly, for the second target area, the electronic device inputs the target research space into each target expert model corresponding to the second target area. The target expert model outputs the sub-target fluid mechanics parameters corresponding to the second target area, and in turn, each target expert model outputs the sub-target fluid mechanics parameters corresponding to each target area.

[0116] Step S2032: Fuse the sub-target fluid mechanics parameters and output the regional fluid mechanics parameters corresponding to the target area.

[0117] In an alternative embodiment, for each target area, when the preset expert model selection network outputs each target expert model corresponding to the target area, it can simultaneously output the weight information corresponding to each target expert model. Then, the electronic device performs a weighted fusion process on the sub-target fluid mechanics parameters output by each target expert model based on the weight information corresponding to each target expert model, and outputs the regional fluid mechanics parameters corresponding to the target area.

[0118] Optionally, for each target area, the electronic device can assign corresponding weights to each target expert model according to factors such as the performance, credibility, or applicability in the target area of each target expert model corresponding to the target area. Then, the electronic device performs a weighted fusion process on the sub-target fluid mechanics parameters output by each target expert model based on the weight information corresponding to each target expert model, and outputs the regional fluid mechanics parameters corresponding to the target area.

[0119] Optionally, the electronic device can also use machine learning algorithms, such as neural networks, support vector machines, etc., to fuse multiple sub-target fluid mechanics parameters. First, a set of training data including sub-target fluid mechanics parameters and corresponding real regional fluid mechanics parameters (obtained through experimental measurement or other reliable methods) needs to be prepared. Then, these data are used to train the machine learning model so that it learns the mapping relationship between the sub-target fluid mechanics parameters and the real regional fluid mechanics parameters. During actual fusion, each sub-target fluid mechanics parameter is input into the trained model, and the fused regional fluid mechanics parameters are output. This method can handle complex parameter relationships and non-linear problems, but requires a large amount of training data and high computing resources.

[0120] Optionally, the electronic device can also establish a comprehensive physical model based on the physical characteristics of the target area and relevant physical laws to fuse the hydrodynamic parameters of sub-targets. For example, in a target area involving multiphase flow, based on the basic equations of multiphase flow and the interaction relationships between phases, the hydrodynamic parameters of each phase calculated by different target expert models can be fused. By solving the comprehensive physical model, the regional hydrodynamic parameters that can more accurately reflect the actual situation of the target area can be obtained.

[0121] Step S2033: Fuse the regional hydrodynamic parameters corresponding to each target area to generate the target hydrodynamic parameters for the target research space.

[0122] Optionally, the electronic device can use an interpolation algorithm to extend the regional hydrodynamic parameters to the entire target research space according to the positions and geometric relationships of each target area in the target research space. For example, for adjacent target areas, linear interpolation or spline interpolation methods can be used to smoothly transition the parameter values at the regional boundaries. When simulating the wind environment in an urban block, given the wind speed and wind direction parameters of each block (target area), the wind speed and wind direction distribution in the open areas between blocks can be obtained through interpolation, thereby generating the wind field parameters for the entire urban area.

[0123] Optionally, if each target area is based on grid division, the electronic device can merge the grids of all target areas into an overall grid, and then map the regional hydrodynamic parameters to the new grid nodes. In this process, it may be necessary to refine or coarsen the grid to ensure the accuracy of the parameters and the calculation efficiency. For example, when calculating the fluid flow under complex terrain, different areas may use grids with different precisions. These grids can be integrated through the grid fusion method, and the parameters of each area can be accurately transferred to the overall grid.

[0124] Optionally, the electronic device can also use machine learning models, such as neural networks, random forests, etc., to fuse the regional hydrodynamic parameters of each target area. First, use the parameters of each area as input features, and at the same time combine other relevant information of the target research space (such as the overall geometric shape, boundary conditions, etc.), and use the true hydrodynamic parameters of the target research space (obtained through experimental measurement or high-precision simulation) as output labels to train the machine learning model. The trained model can predict the target hydrodynamic parameters of the target research space based on the input parameters of each area. For example, use a neural network to fuse the temperature, pressure, and flow rate parameters of multiple target areas to predict the heat flow distribution of the entire target research space.

[0125] Optionally, the electronic device can also establish a relationship model between the parameters of each target region and the overall parameters of the target research space based on statistical methods such as multiple linear regression and principal component analysis. By analyzing and modeling a large amount of known data, the contribution weights of the regional parameters to the overall parameters are determined, and then the regional parameters are fused according to these weights. For example, the main components of the parameters of each target region are extracted through principal component analysis, and then the target hydrodynamic parameters of the target research space are calculated based on the weights of these components.

[0126] The hydrodynamic parameter calculation method provided by the embodiments of the present application divides the target research space into regions to obtain multiple target regions, so as to analyze the characteristics of each target region in more detail. The target regions corresponding to the target research space are input into a preset expert model selection network. For each target region, the preset expert model selection network extracts features of the target region to obtain target parameter features corresponding to the target region. Different target regions have unique hydrodynamic characteristics. The preset expert model selection network extracts features of the target region, which can accurately capture these characteristics and convert them into target parameter features, ensuring the accuracy of the extracted target parameter features. Then, based on the target parameter features, at least one target expert model corresponding to the target region is output, ensuring that each output target expert model matches the target region. Thus, the target expert model can accurately simulate the flow characteristics of a specific target region, greatly improving the calculation accuracy, and can match the complexity of the target region with the target expert model. A model with low computational cost can be used for simple regions, and a high-precision model can be used for complex regions, reasonably allocating computational resources and reducing unnecessary computational volume, thereby improving the computational efficiency. In addition, the situation of the target research space in practical applications is complex and changeable, and a single model is difficult to adapt to the flow characteristics of all regions. The method of dividing regions and selecting models enhances the adaptability of the model. When the boundary conditions or internal structure of the target research space change, for machine learning type target experts, only the model selection of the affected region needs to be adjusted, without having to recalculate the entire space. When simulating the impact of river diversion on the flow field of the surrounding waters, only the model calculation for the diverted region and the affected surrounding regions needs to be reselected, which can quickly respond to changes and adapt to different working conditions.

[0127] Finally, for each target area, input the target research space into each target expert model corresponding to the target area. The target expert model outputs the sub-target hydrodynamic parameters corresponding to the target area. The hydrodynamic characteristics of different target areas vary greatly. Each target expert model simulates the flow characteristics of a specific target area and can more accurately reflect the physical phenomena within the specific target area, thereby ensuring the accuracy of the sub-target hydrodynamic parameters corresponding to the target area. Then, fuse the sub-target hydrodynamic parameters to output the regional hydrodynamic parameters corresponding to the target area. This avoids the accumulation of errors that may occur when a single model simulates the entire target research space in different regions. Within each target area, the calculations of each target expert model are relatively independent, reducing the propagation of errors. Moreover, during the fusion process, comprehensive processing is performed on the sub-target hydrodynamic parameters, further improving the accuracy of the regional hydrodynamic parameters corresponding to the target area. Finally, fuse the regional hydrodynamic parameters corresponding to each target area to generate the target hydrodynamic parameters corresponding to the target research space, ensuring the accuracy of the generated target hydrodynamic parameters.

[0128] According to an embodiment of the present invention, the present invention also provides an embodiment of a method for training a preset expert model selection network. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0129] In this embodiment, a method for training a preset expert model selection network is provided. It is applied to the preset expert model selection network in any of the above embodiments and can be used for the above-mentioned electronic device. Figure 3 It is a flowchart of the method for training a preset expert model selection network according to an embodiment of the present invention. As Figure 3 shown, this process includes the following steps:

[0130] Step S301, obtain a training data set.

[0131] Among them, the training data set includes each training research space and the training hydrodynamic parameters corresponding to each training research space under various working conditions.

[0132] Specifically, the electronic device can receive the training data set input by the user, can also receive the training data set sent by other devices, and can also receive the training data set sent by other devices. The embodiments of the present application do not make specific limitations on the manner in which the electronic device obtains the training data set.

[0133] Step S302: Input each training research space into each training expert model included in the initial expert model selection network, and each training expert model outputs the virtual fluid dynamics parameters corresponding to each training research space under various working conditions.

[0134] Optionally, the electronic device can sequentially input each training research space into each training expert model included in the initial expert model selection network. During the input process, the data needs to be formatted and organized according to the requirements of the training expert model. For some grid-based models, the geometric data of the training research space needs to be converted into grid data; for machine learning models, the data needs to be organized into an appropriate form of feature vectors.

[0135] After receiving the data of the training research space, each training expert model performs calculations according to its own algorithm and principle. For example, the RANS model will calculate virtual fluid dynamics parameters such as velocity field and pressure field by solving the Reynolds-averaged Navier-Stokes equations based on the input geometric shape, boundary conditions, and fluid physical properties, etc.; the ANN model outputs the predicted virtual fluid dynamics parameters through calculations and mappings of multiple layers of neurons on the input data. During the calculation process, the training expert model will consider the influence of various working conditions, such as different flow rates, temperatures, pressures, etc., and output the virtual fluid dynamics parameters corresponding to various working conditions.

[0136] In an optional implementation manner of the present application, the training research space includes at least one training area; each training area is obtained by dividing the training research space by area; the training fluid dynamics parameters include sub-training fluid dynamics parameters corresponding to each training area. The above step S302 may include the following steps:

[0137] Step S3021: Input each training research space into each training expert model included in the initial expert model selection network.

[0138] Specifically, the electronic device inputs each training research space into each training expert model included in the initial expert model selection network.

[0139] Step S3022: For each training area in the training research space, each training expert model extracts features from each training area and outputs the training parameter features corresponding to each training area.

[0140] Specifically, for each training area in the training research space, each training expert model extracts features from each training area according to its own calculation method and outputs the training parameter features corresponding to each training area.

[0141] Exemplarily, for a training area involving fluids, a training expert model (such as a model related to the Navier-Stokes equations, a computational fluid dynamics model, etc.) extracts features based on the basic principles of fluid mechanics. For example, by calculating the velocity gradient, shear stress features can be obtained, which are very important for analyzing the viscous force and flow state of the fluid; by solving the energy equation, temperature distribution features can be obtained for studying the heat transfer process. When simulating the water flow in a certain training area of a river, the model can extract features such as the velocity distribution, pressure distribution, and vorticity in this area, which reflect the motion state and energy conversion of the water flow.

[0142] When there is a heat transfer phenomenon in the training area, a heat transfer model (such as a model related to Fourier's law) extracts features related to the temperature field and heat flow. For example, parameters such as heat flux and temperature gradient are calculated, and these features are crucial for understanding the direction and rate of heat transfer in the area. When simulating the training area near the exterior wall of a building, the heat transfer model can extract the temperature features of the exterior wall surface and the heat flux density features through the exterior wall, providing a basis for evaluating the insulation performance of the building.

[0143] For the training expert model of the neural network, when using a neural network (such as a convolutional neural network CNN, a recurrent neural network RNN, etc.) to extract features, the training expert model automatically learns the patterns and features in the training area data. For data with a spatial structure (such as image data of a fluid velocity field, temperature field), CNN can extract local and global features through convolutional layers and pooling layers. For example, when analyzing the flame image data in a combustion chamber, CNN can extract features such as the shape, brightness, and color of the flame, which are closely related to the combustion state and efficiency. RNN is suitable for processing data with time series features. For example, when monitoring the change of fluid parameters over time in a certain training area, RNN can extract features such as trends and periodicities in the time series.

[0144] Step S3023, based on each training parameter feature, output the sub-virtual fluid mechanics parameters corresponding to each training area under various working conditions.

[0145] Specifically, each training expert model, based on each training parameter feature, outputs the sub-virtual fluid mechanics parameters corresponding to each training area under various working conditions.

[0146] Step S303, based on the correspondence between each virtual fluid mechanics parameter and each training fluid mechanics parameter, adjust the parameters of the initial expert model selection network to obtain a preset expert model selection network.

[0147] Specifically, the above step S303 may include the following steps:

[0148] Step S3031: For each training area, based on the attribute information corresponding to each training area, determine at least one evaluation mechanism corresponding to each training area.

[0149] Among them, the attribute information corresponding to the training area can be attribute information such as physical attributes, flow characteristic attributes, material attributes, and external environment attributes. The embodiments of the present application do not make specific limitations on the attribute information corresponding to the training area. Among them, physical attributes can include geometric features of the training area, such as shape, size, boundary conditions, etc., and physical properties of the fluid, such as density, viscosity, compressibility, etc. For example, in the training area of studying fluid flow in a pipeline, geometric attributes such as the diameter, length, and bending degree of the pipeline, and density and viscosity attributes corresponding to whether the fluid is water or oil, are all important physical attribute information. Flow characteristic attributes can include velocity distribution, pressure distribution, temperature distribution, turbulence intensity, etc. of the fluid in the training area. For example, when simulating the training area of the atmospheric boundary layer, the wind speed and direction changes at different heights, and the resulting pressure difference and temperature gradient, are all key attributes reflecting the flow characteristics. Material attributes can include, for example, the thermal conductivity, roughness, elasticity, etc. of the material. When studying fluid flow in a heat exchanger, the thermal conductivity of the heat exchanger pipe material and the roughness of the inner wall will have an important impact on the heat transfer and flow of the fluid. External environment attributes can include, for example, environmental temperature, pressure, gravitational field, etc. When studying ocean currents in a certain training area, external environment attributes such as the environmental temperature, salinity of seawater, and water pressure corresponding to the depth where it is located will affect the flow state of the ocean currents.

[0150] Among them, the evaluation mechanism can include a physical evaluation mechanism (such as Reynolds number performance, Mach number performance, turbulence intensity performance, etc.), a prediction accuracy evaluation mechanism (such as key physical quantity error, integral characteristic error (such as lift and drag)), a calculation efficiency evaluation mechanism (such as calculation duration, memory usage), and can also include other evaluation mechanisms. The embodiments of the present application do not make specific limitations on the evaluation mechanism.

[0151] For each training area, the electronic device determines at least one evaluation mechanism corresponding to each training area based on the attribute information corresponding to each training area.

[0152] Step S3032: Based on the objective function corresponding to each evaluation mechanism, calculate the evaluation scores corresponding to each training expert model under various working conditions.

[0153] Specifically, the electronic device can construct the corresponding objective function according to the evaluation mechanism of each determined training area. For example, if the evaluation mechanism is based on efficiency evaluation, the objective function can be set to calculate the efficiency value of energy conversion or transmission in the training area. For example, the combustion efficiency of an engine combustion chamber can be expressed as the ratio of the output effective energy to the input total energy.

[0154] When there are multiple evaluation mechanisms, the electronic device can perform a weighted combination of the sub-objective functions corresponding to each evaluation mechanism to form a comprehensive objective function. Among them, the determination of the weights of the sub-objective functions corresponding to each evaluation mechanism needs to be determined according to the requirements of the specific problem and the importance of each evaluation mechanism. For example, in some cases, flow stability may be more critical, and a higher weight can be assigned to it.

[0155] Then, substitute the sub-virtual hydrodynamics parameters output by each training expert model for each training area under various working conditions and the sub-training hydrodynamics parameters corresponding to each training area into the corresponding objective function for calculation. This may involve various mathematical operations, such as addition, subtraction, multiplication, division, integration, differentiation, etc. For some complex objective functions, numerical calculation methods may be required to solve them. For example, when calculating the objective function based on stability evaluation in a turbulent flow field, operations such as fast Fourier transform (FFT) on velocity pulsation data may be required to analyze its spectral characteristics and stability indicators.

[0156] Step S3033, determine at least one candidate expert model according to the evaluation scores of each training expert model under various working conditions.

[0157] Optionally, for each training area under various working conditions, the electronic device can compare the evaluation scores corresponding to the training expert models and select each candidate expert model with the top N evaluation scores.

[0158] Optionally, for each training area under various working conditions, the electronic device can also compare the evaluation scores corresponding to the training expert models with a preset scoring threshold and select each candidate expert model with an evaluation score greater than the preset scoring threshold.

[0159] Optionally, for each training area under various working conditions, the initial expert model selection network can also determine the activation probability corresponding to each training expert model based on the evaluation scores output by each training expert model, and then determine the training expert models with activation probabilities greater than the preset threshold as candidate expert models.

[0160] Step S3034, based on each candidate expert model, adjust the parameters of the initial expert model selection network to obtain a preset expert model selection network.

[0161] Specifically, the above step S3034 may include the following steps:

[0162] Step b1, for each training area, perform a fusion process on the sub-virtual hydrodynamics parameters output by each candidate expert model corresponding to the training area to obtain the first fusion hydrodynamics parameter.

[0163] Optionally, for each training region, when the initial expert model selection network outputs each candidate expert model corresponding to the training region, it can simultaneously output the activation probability corresponding to each candidate expert model. Based on the activation probability corresponding to each candidate expert model, the weight information corresponding to each candidate expert model is determined. Then, the electronic device performs weighted fusion processing on each sub virtual hydrodynamics parameter output by each candidate expert model based on the weight information corresponding to each candidate expert model, and outputs the first fusion hydrodynamics parameter corresponding to the training region.

[0164] Optionally, for each training region, the initial expert model selection network can determine the weights assigned to each candidate expert model according to the evaluation scores corresponding to each candidate expert model in the training region. Then, the electronic device performs weighted fusion processing on each sub virtual hydrodynamics parameter output by each candidate expert model based on the weight information corresponding to each candidate expert model, and outputs the first fusion hydrodynamics parameter corresponding to the training region.

[0165] Optionally, the initial expert model selection network can also use machine learning algorithms, such as neural networks, support vector machines, etc., to fuse multiple sub virtual hydrodynamics parameters. First, a set of training data containing sub virtual hydrodynamics parameters and corresponding true first fusion hydrodynamics parameters (obtained through experimental measurement or other reliable methods) needs to be prepared. Then, these data are used to train the machine learning model so that it learns the mapping relationship between the sub virtual hydrodynamics parameters and the true first fusion hydrodynamics parameters. During actual fusion, each sub virtual hydrodynamics parameter is input into the trained model, and the fused first fusion hydrodynamics parameter is output. This method can handle complex parameter relationships and nonlinear problems, but requires a large amount of training data and high computing resources.

[0166] Optionally, the initial expert model selection network can also establish a comprehensive physical model to fuse sub virtual hydrodynamics parameters according to the physical characteristics of the training region and relevant physical laws. For example, in a training region involving multiphase flow, based on the basic equations of multiphase flow and the interaction relationships between phases, the hydrodynamics parameters of each phase calculated by different candidate expert models can be fused. By solving the comprehensive physical model, a first fusion hydrodynamics parameter that more accurately reflects the actual situation of the training region can be obtained.

[0167] Step b2, calculate the loss function between the first fusion hydrodynamics parameter corresponding to each training region and the corresponding sub-training hydrodynamics parameter.

[0168] Specifically, the initial expert model selection network can also calculate the loss function between the first fusion hydrodynamics parameter corresponding to each training region and the corresponding sub-training hydrodynamics parameter.

[0169] Among them, the loss function can be the Mean Squared Error (MSE), the Mean Absolute Error (MAE), the Huber loss function, or other loss functions. The embodiments of the present application do not make specific limitations on the loss function.

[0170] Step b3: Fuse the loss functions corresponding to each training region to generate a target loss function.

[0171] Optionally, the initial expert model selection network can determine the possible different degrees of influence of different training regions on the overall model performance according to the characteristics of each training region in the training research space, such as the size of the region, the complexity of the physical process, the importance in the entire training research space, etc., so as to determine the weight information of the loss functions corresponding to each training region. For example, when simulating the internal flow field of a large building, public areas with frequent human activities (such as lobbies) may have a greater impact on comfort and ventilation effects than some auxiliary rooms (such as storage rooms). Therefore, when fusing the loss functions, the loss function of the public area should be given higher attention.

[0172] Then, the initial expert model selection network fuses the loss functions corresponding to each training region according to the weight information of the loss functions corresponding to each training region to generate a target loss function.

[0173] Optionally, the initial expert model selection network can also learn the relationship between the loss functions of each training region based on machine learning algorithms (such as neural networks, random forests, etc.) and automatically generate a target loss function. By preparing a set of training data including the loss functions of each training region and known comprehensive evaluation indicators (such as the overall model performance score), training the machine learning model so that it can output an appropriate target loss function value according to the input loss functions of each training region.

[0174] Step b4: Adjust the parameters of the initial expert model selection network according to the target loss function until the target loss function stabilizes within a preset region to obtain a preset expert model selection network.

[0175] Specifically, the electronic device can determine the initial settings for parameter adjustment such as the initial learning rate and the upper limit of the number of iterations. Among them, the selection of the learning rate is crucial. An overly large learning rate may cause the parameters to skip the optimal solution during the update, while an overly small learning rate will make the training process too slow. The upper limit of the number of iterations is used to control the maximum number of iterations in the training process to prevent the algorithm from falling into an infinite loop.

[0176] Then, according to the selected parameter adjustment method, calculate the gradient of the objective loss function with respect to the network parameters in each iteration and update the parameters. During the process of updating the parameters, the value of the objective loss function after each iteration can be recorded to observe the changing trend of the loss function. During the iteration process, monitor the change of the objective loss function. If it is found that the loss function decreases too slowly, the learning rate can be appropriately increased; if the loss function fluctuates or increases, it indicates that the learning rate may be too large and needs to be decreased. Some adaptive learning rate algorithms (such as Adagrad, Adam, etc.) can automatically adjust the learning rate, simplifying this process.

[0177] Preset a stable region of the objective loss function, which can usually be represented by setting a threshold or a range. When the change amount of the objective loss function in consecutive several iterations is less than a certain threshold, or the value of the objective loss function stabilizes within the preset region, it is considered that the parameter adjustment process reaches a stable state. For example, it is considered to reach stability when the change amount of the objective loss function in consecutive 10 iterations is less than 0.001.

[0178] When the objective loss function meets the stable condition, the initial expert model selection network after parameter adjustment at this time becomes the preset expert model selection network. The parameters of this network have been optimized and can accurately select a suitable expert model to a certain extent to better handle the problems in the target research space.

[0179] The preset expert model selection network training method provided by the embodiments of this application obtains a training data set. The training data set covers the true hydrodynamic parameters of various training research spaces under different working conditions. Input each training research space into each training expert model included in the initial expert model selection network. For each training region in the training research space, each training expert model extracts features of each training region and outputs the training parameter features corresponding to each training region. Thus, each training expert model can capture the accurate features corresponding to the training region for each training region, ensuring the accuracy of the output training parameter features. In addition, the regional processing makes the training of each training region relatively independent, avoiding the interference of complex interactions between different training regions on the training results. Then, based on each training parameter feature, output the sub-virtual hydrodynamic parameters corresponding to each training region under various working conditions, ensuring the accuracy of the sub-virtual hydrodynamic parameters corresponding to each training region under various working conditions. The above method processes each training region separately, and can reasonably allocate computing resources according to the complexity and computing requirements of each region. In addition, the regional training can be carried out in parallel, using multi-core computing resources to train multiple training regions simultaneously, improving the training efficiency.

[0180] Then, for each training region, based on the attribute information corresponding to each training region, at least one evaluation mechanism corresponding to each training region is determined, so as to more accurately measure the performance of the training expert model in a specific training region. Based on the objective functions corresponding to the evaluation mechanisms, the evaluation scores corresponding to each training expert model under various working conditions are calculated, so as to comprehensively evaluate the performance of each training expert model in each training region under different working conditions. According to the evaluation scores corresponding to each training expert model under various working conditions, at least one candidate expert model is determined therefrom, so that the parameters of the initial expert model selection network are adjusted based on the candidate expert model, making the initial expert model selection network more inclined to select training expert models with excellent performance. For each training region, the sub-virtual hydrodynamics parameters output by each candidate expert model corresponding to the training region are fused to obtain the first fused hydrodynamics parameter, which can integrate the advantages of multiple candidate expert models and make up for the limitations of a single candidate expert model, ensuring the accuracy of the obtained first fused hydrodynamics parameter. The loss function between the first fused hydrodynamics parameter corresponding to each training region and the corresponding sub-training hydrodynamics parameter is calculated, and the loss functions corresponding to each training region are fused to generate the target loss function, providing a clear optimization target for the parameter adjustment of the initial expert model selection network. The parameters of the initial expert model selection network are adjusted according to the target loss function, enabling the initial expert model selection network to gradually learn how to select and combine models to minimize the difference between the prediction result and the true value. As the parameters are continuously adjusted, the prediction accuracy of the preset expert model selection network will gradually increase, and finally it can more accurately simulate the hydrodynamics characteristics of different training regions. In addition, the network parameters are adjusted based on the loss functions of each training region, enabling the preset expert model selection network to better adapt to the characteristics of different training regions and the changes in various working conditions. Each training region has its unique hydrodynamics characteristics and boundary conditions. By separately calculating and fusing the loss functions, the preset expert model selection network can perform personalized optimization for different regions, improving the adaptability of the preset expert model selection network in different regions. Considering the changes in hydrodynamics phenomena under different working conditions, this method can also enable the preset expert model selection network to maintain good performance under various working conditions, enhancing the reliability of the preset expert model selection network.

[0181] In this embodiment, a hydrodynamics parameter calculation device is further provided. This device is used to implement the above-mentioned embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0182] This embodiment provides a hydrodynamics parameter calculation device, asFigure 4 As shown in the figure, it includes:

[0183] A first acquisition module 401, configured to acquire a target research space;

[0184] An output module 402, configured to input the target research space into a preset expert model selection network, and output at least one target expert model corresponding to the target research space;

[0185] A calculation module 403, configured to have each target expert model identify the target research space, and calculate target hydrodynamic parameters corresponding to the target research space according to the identification result.

[0186] In some alternative embodiments, the output module 402 is specifically configured to divide the target research space by region to obtain a plurality of target regions; input each target region corresponding to the target research space into the preset expert model selection network; for each target region, output at least one target expert model corresponding to the target region.

[0187] In some alternative embodiments, the output module 402 is specifically configured to, for each target region, the preset expert model selection network extracts features of the target region to obtain target parameter features corresponding to the target region; based on the target parameter features, output at least one target expert model corresponding to the target region.

[0188] In some alternative embodiments, the calculation module 403 is specifically configured to, for each target region, input the target research space into each target expert model corresponding to the target region, and the target expert model outputs sub-target hydrodynamic parameters corresponding to the target region; fuse the sub-target hydrodynamic parameters to output regional hydrodynamic parameters corresponding to the target region; fuse the regional hydrodynamic parameters corresponding to each target region to generate target hydrodynamic parameters corresponding to the target research space.

[0189] In this embodiment, a device for training a preset expert model selection network is further provided. This device is used to implement the above embodiments and preferred implementation manners, and those that have been described will not be repeated here. As used hereinafter, the term "module" can be a combination of software and / or hardware that can implement a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0190] This embodiment provides a device for training a preset expert model selection network, which is applied to the preset expert model selection network of any one of the above embodiments, as Figure 5 shown in the figure, including:

[0191] A second acquisition module 501, configured to acquire a training data set, where the training data set includes each training research space and the corresponding training hydrodynamics parameters of each training research space under various working conditions;

[0192] An input module 502, configured to input each training research space into each training expert model included in the initial expert model selection network, and each training expert model outputs the corresponding virtual hydrodynamics parameters of each training research space under various working conditions;

[0193] An adjustment module 503, configured to adjust the parameters of the initial expert model selection network based on the corresponding relationship between each virtual hydrodynamics parameter and each training hydrodynamics parameter, so as to obtain a preset expert model selection network.

[0194] In some optional embodiments, the training research space includes at least one training area; each training area is obtained by dividing the training research space by area; the training hydrodynamics parameters include the sub-training hydrodynamics parameters corresponding to each training area; the input module 502 is specifically configured to input each training research space into each training expert model included in the initial expert model selection network; for each training area in the training research space, each training expert model extracts features of each training area and outputs the training parameter features corresponding to each training area; based on each training parameter feature, the sub-virtual hydrodynamics parameters corresponding to each training area under various working conditions are output.

[0195] In some optional embodiments, the adjustment module 503 is specifically configured to, for each training area, determine at least one evaluation mechanism corresponding to each training area based on the attribute information corresponding to each training area; calculate the evaluation scores corresponding to each training expert model under various working conditions based on the objective function corresponding to each evaluation mechanism; determine at least one candidate expert model therefrom according to the evaluation scores corresponding to each training expert model under various working conditions; and adjust the parameters of the initial expert model selection network based on each candidate expert model to obtain a preset expert model selection network.

[0196] In some optional embodiments, the adjustment module 503 is specifically configured to, for each training area, perform fusion processing on the sub-virtual hydrodynamics parameters output by each candidate expert model corresponding to the training area to obtain a first fusion hydrodynamics parameter; calculate the loss function between the first fusion hydrodynamics parameter corresponding to each training area and the corresponding sub-training hydrodynamics parameter; perform fusion processing on the loss functions corresponding to each training area to generate an objective loss function; and adjust the parameters of the initial expert model selection network according to the objective loss function until the objective loss function is stable within a preset region, so as to obtain a preset expert model selection network.

[0197] The further function descriptions of the above-mentioned various modules and units are the same as those in the corresponding above embodiments, and will not be repeated here.

[0198] The hydrodynamic parameter calculation device and the preset expert model selection network training device in this embodiment are presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0199] The embodiment of the present invention also provides an electronic device having the above Figure 4 shown hydrodynamic parameter calculation device and Figure 5 shown preset expert model selection network training device.

[0200] Please refer to Figure 6 , Figure 6 which is a schematic structural diagram of an electronic device provided by an alternative embodiment of the present invention. As Figure 6 shown, the electronic device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the electronic device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple electronic devices can be connected, and each device provides some necessary operations (such as a server array, a set of blade servers, or a multi-processor system). Figure 6 One processor 10 is taken as an example in

[0201] The processor 10 can be a central processor, a network processor, or a combination thereof. Among them, the processor 10 can further include a hardware integrated circuit. The above hardware integrated circuit can be an application specific integrated circuit, a programmable logic device, or a combination thereof. The above programmable logic device can be a complex programmable logic device, a field programmable gate array, a generic array logic, or any combination thereof.

[0202] Among them, the memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiments.

[0203] The memory 20 may include a program storage area and a data storage area. The program storage area may store an operating system and application programs required for at least one function. The data storage area may store data created according to the use of the electronic device and the like. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 may optionally include a memory remotely provided with respect to the processor 10, and these remote memories may be connected to the electronic device through a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0204] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 may further include a combination of the above types of memories.

[0205] The electronic device further includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30, and the output device 40 may be connected through a bus or other means. Figure 6 Taking connection through a bus as an example.

[0206] The input device 30 may receive input digital or character information, and generate key signal inputs related to the user settings and function controls of the electronic device, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointing stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED), and a haptic feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes, but is not limited to, a liquid crystal display, a light-emitting diode, a display, and a plasma display. In some alternative embodiments, the display device may be a touch screen.

[0207] Embodiments of the present invention also provide a computer-readable storage medium. The method according to the embodiments of the present invention can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code that is originally stored in a remote storage medium or a non-transitory machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored as such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0208] A part of the present invention can be applied as a computer program product, for example, computer program instructions, which when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should be able to understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Herein, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to the computer.

[0209] Although the embodiments of the present invention are described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for calculating hydrodynamic parameters, characterized in that, The method includes: Obtain a target research space; Input the target research space into a preset expert model selection network, and output at least one target expert model corresponding to the target research space; Each of the target expert models identifies the target research space, and calculates target hydrodynamic parameters corresponding to the target research space according to the identification results.

2. The method according to claim 1, wherein The step of inputting the target research space into a preset expert model selection network and outputting at least one target expert model corresponding to the target research space includes: Divide the target research space by region to obtain a plurality of target regions; Input each of the target regions corresponding to the target research space into the preset expert model selection network; For each of the target regions, output at least one of the target expert models corresponding to the target region.

3. The method according to claim 2, wherein The step of, for each of the target regions, outputting at least one of the target expert models corresponding to the target region includes: For each of the target regions, the preset expert model selection network extracts features of the target region to obtain target parameter features corresponding to the target region; Based on the target parameter features, output at least one of the target expert models corresponding to the target region.

4. The method according to claim 2, wherein The step of each of the target expert models identifying the target research space and calculating target hydrodynamic parameters corresponding to the target research space according to the identification results includes: For each of the target regions, input the target research space into each of the target expert models corresponding to the target region, and the target expert model outputs sub-target hydrodynamic parameters corresponding to the target region; Fuse the sub-target hydrodynamic parameters to output regional hydrodynamic parameters corresponding to the target region; Fuse the regional hydrodynamic parameters corresponding to each of the target regions to generate the target hydrodynamic parameters corresponding to the target research space.

5. A method for training a preset expert model selection network, characterized in that, When applied to the preset expert model selection network according to any one of claims 1-4, the method includes: Obtain a training data set, where the training data set includes each training research space and training hydrodynamic parameters corresponding to each training research space under various working conditions; Input each of the training research spaces into each of the training expert models included in the initial expert model selection network, and each of the training expert models outputs virtual hydrodynamic parameters corresponding to each training research space under various working conditions; Based on the correspondence between the virtual hydrodynamic parameters and the training hydrodynamic parameters, adjust the parameters of the initial expert model selection network to obtain the preset expert model selection network.

6. The method according to claim 5, characterized in that, The training research space includes at least one training region; each of the training regions is obtained by dividing the training research space by region; the training hydrodynamic parameters include sub-training hydrodynamic parameters corresponding to each of the training regions; the step of inputting each of the training research spaces into each of the training expert models included in the initial expert model selection network, and each of the training expert models outputs virtual hydrodynamic parameters corresponding to each training research space under various working conditions includes: Input each of the training research spaces into each of the training expert models included in the initial expert model selection network; For each of the training regions in the training research space, each of the training expert models extracts features from each of the training regions and outputs training parameter features corresponding to each of the training regions; Based on each of the training parameter features, output sub-virtual fluid mechanics parameters corresponding to each of the training regions under various working conditions.

7. The method according to claim 6, wherein The parameter adjustment of the initial expert model selection network based on the corresponding relationship between each of the virtual fluid mechanics parameters and each of the training fluid mechanics parameters to obtain the preset expert model selection network includes: For each of the training regions, based on the attribute information corresponding to each of the training regions, determine at least one evaluation mechanism corresponding to each of the training regions; Based on the objective functions corresponding to each of the evaluation mechanisms, calculate the evaluation scores corresponding to each of the training expert models under various working conditions; According to the evaluation scores corresponding to each of the training expert models under various working conditions, determine at least one candidate expert model therefrom; Based on each of the candidate expert models, perform parameter adjustment on the initial expert model selection network to obtain the preset expert model selection network.

8. The method according to claim 7, wherein The parameter adjustment of the initial expert model selection network based on each of the candidate expert models to obtain the preset expert model selection network includes: For each of the training regions, perform fusion processing on the sub-virtual fluid mechanics parameters output by each of the candidate expert models corresponding to the training region to obtain a first fusion fluid mechanics parameter; Calculate the loss function between the first fusion fluid mechanics parameter corresponding to each of the training regions and the corresponding sub-training fluid mechanics parameter; Perform fusion processing on the loss functions corresponding to each of the training regions to generate an objective loss function; According to the objective loss function, perform parameter adjustment on the initial expert model selection network until the objective loss function is stable within a preset region to obtain the preset expert model selection network.

9. A hydrodynamic parameter calculation device, characterized in that, The device includes: A first acquisition module, configured to acquire a target research space; An output module, configured to input the target research space into a preset expert model selection network and output at least one target expert model corresponding to the target research space; A calculation module, configured to each of the target expert models identify the target research space and calculate target fluid mechanics parameters corresponding to the target research space according to the identification result.

10. A training device for a preset expert model selection network, characterized in that Applied to the preset expert model selection network according to any one of claims 1-4, the device includes: A second acquisition module, configured to acquire a training data set, where the training data set includes each training research space and training fluid mechanics parameters corresponding to each of the training research spaces under various working conditions; An input module, configured to input each of the training research spaces into each of the training expert models included in the initial expert model selection network, and each of the training expert models outputs virtual fluid mechanics parameters corresponding to each of the training research spaces under various working conditions; An adjustment module, configured to adjust the parameters of the initial expert model selection network based on the correspondence between each of the virtual fluid dynamics parameters and each of the training fluid dynamics parameters, so as to obtain the preset expert model selection network.

Images